UDCAI/Z-Image-Fun-Distill-ComfyUI

Instalación y uso con Diffusers:

pip install -U diffusers transformers accelerate

import torch
from diffusers import DiffusionPipeline

# switch to "mps" for apple devices
pipe = DiffusionPipeline.from_pretrained("Tongyi-MAI/Z-Image", dtype=torch.bfloat16, device_map="cuda")
pipe.load_lora_weights("UDCAI/Z-Image-Fun-Distill-ComfyUI")

prompt = "A realistic photo of a young woman with short black hair and vibrant pink tips, wearing a metallic spiked headband and a matching spiked choker. She has large, expressive red eyes and is depicted with a slightly surprised or breathless expression, her mouth parted. She is dressed in a black and white cheerleader uniform, including a cropped tank top with a bold graphic print that says \"PHAETHON\" and a pleated mini-skirt. She is posing with her hands holding yellow pom-poms. Her skin is glistening with droplets of sweat or water, and there are floating water bubbles around her. Attached to her backside is a large, dark grey shark-like tail fin decorated with white graffiti. The background shows a sunlit athletic stadium with a running track under a clear blue sky. The overall style is high-detail with vibrant colors and dynamic lighting."
image = pipe(prompt).images[0]

Recomendaciones del autor: usar con Z-Image, preferiblemente a 8 pasos y CFG 1. Evitar sageattention. Los samplers que añaden ruido por paso, como LCM, parecen funcionar bien.